{
  "family": "higgs_audio_tts",
  "display_name": "Higgs Audio v3 TTS",
  "description": "Boson AI conversational TTS model for expressive speech across 100+ languages, zero-shot voice cloning, and inline control over emotion, style, prosody, pauses, and sound effects.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "100+ languages"
  ],
  "options": {
    "request": [
      {
        "name": "stream_frames_per_event",
        "type": "int",
        "description": "Streaming codec frames per event; grow starts at one and increases to this cap. Each frame is 40 ms of audio.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "stream_chunk_policy",
        "type": "enum",
        "description": "Streaming chunk schedule. The non-causal codec always retains ten frames of left and right context.",
        "required": false,
        "values": [
          "grow",
          "fixed"
        ],
        "default": "grow"
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated AR tokens; zero uses the default. Must not exceed the configured context length.",
        "required": false,
        "min": 0,
        "default": 2048
      },
      {
        "name": "allow_incomplete_generation",
        "type": "bool",
        "description": "Return generated audio with a warning instead of failing when max_tokens is reached before EOC. Audio may be truncated or contain silence/noise.",
        "required": false,
        "default": false
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "AR sampling temperature. Must be positive.",
        "required": false,
        "default": 0.8
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "AR nucleus sampling probability. Must be positive.",
        "required": false,
        "default": 0.8
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "AR top-k sampling limit.",
        "required": false,
        "default": 30,
        "min": 0
      },
      {
        "name": "filter_nan_logits",
        "type": "bool",
        "description": "Filter NaN and infinity during greedy and stochastic top-k selection. All-invalid logits still fail.",
        "required": false,
        "default": false
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Accepted for Python API compatibility; audio sampling does not consume this value. Compatibility-only, but must be finite and positive.",
        "required": false,
        "default": 1.1
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Torch sampling seed; omitted chooses a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "reference_text",
        "type": "string",
        "description": "Transcript accompanying reference audio.",
        "required": false,
        "default": ""
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Framework long-form text chunk size.",
        "required": false,
        "default": 1024,
        "min": 1
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Framework text chunking mode.",
        "required": false,
        "values": [
          "default",
          "tag_aware",
          "japanese",
          "endline"
        ],
        "default": "default"
      }
    ],
    "session": [
      {
        "name": "reference_prefill_mode",
        "type": "enum",
        "description": "Cached-reference prefill attention. auto preserves existing behavior. Try exact if repeated voice-cloning requests produce excessive repetition or unexpectedly long, mostly silent output. May increase prefill time; not a general generation-quality fix. Full-prompt prefill and AR decoding are unchanged.",
        "required": false,
        "values": [
          "auto",
          "exact"
        ],
        "default": "auto"
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Weight storage type.",
        "required": false,
        "preset": "weight_type_full",
        "default": "native"
      },
      {
        "name": "ar_weight_type",
        "type": "enum",
        "description": "AR weight storage; inherits weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "codec_weight_type",
        "type": "enum",
        "description": "Codec weight storage; inherits weight_type.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "ar_weight_context_mb",
        "type": "int",
        "description": "AR weight context in MiB; 4096 on 64-bit builds, 1024 on 32-bit builds.",
        "required": false,
        "min": 1
      },
      {
        "name": "codec_weight_context_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 1536
      },
      {
        "name": "ar_decode_graph_arena_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "codec_decode_graph_arena_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "codec_encode_graph_arena_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "reference_cache_slots",
        "type": "int",
        "description": "Encoded reference cache slots; zero disables reuse.",
        "required": false,
        "min": 0,
        "default": 1
      },
      {
        "name": "attention",
        "type": "enum",
        "description": "Attention lowering preference.",
        "required": false,
        "values": [
          "auto",
          "flash",
          "eager"
        ],
        "default": "auto"
      }
    ],
    "load": []
  },
  "capabilities": {
    "clone": [
      "speaker_reference"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "higgs_audio_tts_4b_q8_0",
    "tags": [
      "TTS",
      "Clone",
      "GGUF"
    ],
    "docs": [
      "docs/models/higgs_audio_tts.md",
      "docs/tts.md",
      "docs/gguf.md"
    ]
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "higgs_audio_tts_4b_q8_0",
      "display_name": "Higgs Audio v3 TTS 4B Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Higgs-Audio-v3-TTS-4B-GGUF",
      "files": [
        "Higgs-Audio-v3-TTS-4B-GGUF/higgs-audio-v3-tts-4b-q8_0.gguf"
      ],
      "strip_prefix": "Higgs-Audio-v3-TTS-4B-GGUF"
    },
    {
      "id": "higgs_audio_tts_4b_bf16",
      "display_name": "Higgs Audio v3 TTS 4B BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Higgs-Audio-v3-TTS-4B-GGUF",
      "files": [
        "Higgs-Audio-v3-TTS-4B-GGUF/higgs-audio-v3-tts-4b-bf16.gguf"
      ],
      "strip_prefix": "Higgs-Audio-v3-TTS-4B-GGUF"
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "chat_template": "model:chat_template.jinja"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_json": "model:tokenizer.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "chat_template": "model:chat_template.jinja"
      },
      "tensors": {
        "weights": "model:model.safetensors.index.json"
      }
    }
  ]
}